Net Benefit analysis_How do people feel about AI.Rmd

---
  title: "Net benefit analysis"
output: html_document
date: "2023-01-31"
---
  
```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

```{r}
# setwd("..") # Use this line to set the working directory on your machine where the data is located. 
```

```{r}
install.packages("tidyverse")
install.packages("ggplot2")
install.packages("haven")
install.packages("scales")
install.packages("readr")
install.packages("data.table")
install.packages("epiDisplay")
```


```{r}
library(tidyverse)
library(ggplot2)
library(haven)
library(scales)
library(readr)
library(data.table)
library(epiDisplay)
```

Read raw data
```{r}
df1 <- read_dta("262400371_Public Attitudes to AI_client version.dta")

# Prep data
# Remove ppt identifying as 'other' and remove 16 year old 
df1 <- df1 %>%
  filter(is.na(RS_Sex) | RS_Sex != 3) %>%
  filter(CS_Age != "16")
```




Transform to data.table format
```{r}
df1_dt <- as.data.table(df1)
```


Now we will re-code all the variables of interest. We need to consider the "DK" and "Prefer not to say" as missing, in order to compute the net benefit. 
For starters, let's list the variable names (for benefits) so we are on the same page:

BENA_MP: Face recognition unblock tech
BENA_BC: Face recognition border control
BENA_PS: Face recognition politicng
BENA_WEL: Eligibility welfare benefits
BENA_JOB: Elejibility jobs
BENA_CAN: Predict risk cancer
BENA_LOAN: Predict risk no pay loan
BENA_SMC: Social media ads consumer products
BENA_SMP: Social media ads political parties
BENA_VASS: Virtual speakers
BENA_VAH: Healthcare virtual assistant
BENA_RVC: Robotic vacuumn cleaner
BENA_RCA: Robotic care assistant
BENA_DC: Driverless cars
BENA_AW: Autonomous weapons
BENA_CCR: AI simulation for climate change research
BENA_VR: VR education

CONA_* is the same, but about the concerns. 
```{r}
## We copy the dataset

df1_dt_1 = copy(df1_dt)     
df1_dt_1 <- as.data.table(df1_dt)

## We create a list with each AI application identifier

list_variables <- c("MP", "BC", "PS", "WEL", "JOB", "CAN", "LOAN", "SMC", "SMP",
                    "VASS", "VAH", "RVC", "RCA", "DC", "AW", "CCR", "VR")


## Now we can re-code the variables for benefits and concerns, in order to add
## missing data and flip the scale.


for(i in (list_variables)) {
  BENA <- paste0('BENA_',i)
  CONA <- paste0('CONA_',i)
  df1_dt_1[get(BENA)== 6, (BENA):= NA] 
  df1_dt_1[get(BENA)== 5, (BENA):= NA]
  df1_dt_1[get(BENA)== 4, (BENA):= 0] 
  df1_dt_1[get(BENA)== 3, (BENA):= -1] 
  df1_dt_1[get(BENA)== 2, (BENA):= -2] 
  df1_dt_1[get(BENA)== 1, (BENA):= -3] 
  df1_dt_1[get(BENA)== -1, (BENA):= 1] 
  df1_dt_1[get(BENA)== -2, (BENA):= 2] 
  df1_dt_1[get(BENA)== -3, (BENA):= 3] 
  df1_dt_1[get(CONA)== 6, (CONA):= NA] 
  df1_dt_1[get(CONA)== 5, (CONA):= NA]
  df1_dt_1[get(CONA)== 4, (CONA):= 0] 
  df1_dt_1[get(CONA)== 3, (CONA):= -1] 
  df1_dt_1[get(CONA)== 2, (CONA):= -2] 
  df1_dt_1[get(CONA)== 1, (CONA):= -3] 
  df1_dt_1[get(CONA)== -1, (CONA):= 1] 
  df1_dt_1[get(CONA)== -2, (CONA):= 2] 
  df1_dt_1[get(CONA)== -3, (CONA):= 3] 
}


## We also need to re-code the variables for AWARENESS

for(i in (list_variables)) {
  AWARE <- paste0('AWARE_',i)
  df1_dt_1[get(AWARE)== 4, (AWARE):= NA] 
  df1_dt_1[get(AWARE)== 2, (AWARE):= 0] 
  df1_dt_1[get(AWARE)== 3, (AWARE):= 0] 
}


## And EXPERIENCE


list_variables_2 <- c("MP", "BC", "PS", "WEL", "JOB", "CAN", "LOAN", "SMC", "SMP",
                    "VASS", "VAH", "RVC", "VR")

for(i in (list_variables_2)) {
  EXP <- paste0('EXP_',i)
  df1_dt_1[get(EXP)== 5, (EXP):= NA] 
  df1_dt_1[get(EXP)== 4, (EXP):= NA]
  df1_dt_1[get(EXP)== 2, (EXP):= 1] 
  df1_dt_1[get(EXP)== 3, (EXP):= 0] 
}



```

Now, following the same approach, we can create the net benefit variables
```{r}
for(i in (list_variables)) {
  BENA <- paste0('BENA_',i)
  CONA <- paste0('CONA_',i)
  NETBEN <- paste0('NETBEN_',i)
  df1_dt_1[is.na(get(BENA))==F & is.na(get(CONA))==F, (NETBEN) := get(BENA)-get(CONA)] 
}


df1_dt_1[, tab1(NETBEN_CCR, graph=F)] ##2320 NA

#Double check in several ways
df1_dt_1[is.na(BENA_CCR)==T | is.na(CONA_CCR)==T, tab1(NETBEN_CCR, graph=F)]
df1_dt_1[, tab1(BENA_CCR-CONA_CCR, graph=F)]

df1_dt_1[, tab1(NETBEN_VASS, graph=F)] ##2179

#Double check in several ways
df1_dt_1[is.na(BENA_VASS)==T | is.na(CONA_VASS)==T, tab1(NETBEN_VASS, graph=F)]
df1_dt_1[, tab1(BENA_VASS-CONA_VASS, graph=F)]

```

We can look now at whethet the mean values make sense for each of the technologies
```{r}
df1_dt_1[, summ(NETBEN_MP, graph=F)] 
df1_dt_1[, summ(NETBEN_BC, graph=F)] 
df1_dt_1[, summ(NETBEN_PS, graph=F)] 
df1_dt_1[, summ(NETBEN_WEL, graph=F)] 
df1_dt_1[, summ(NETBEN_JOB, graph=F)] 
df1_dt_1[, summ(NETBEN_CAN, graph=F)] 
df1_dt_1[, summ(NETBEN_LOAN, graph=F)] 
df1_dt_1[, summ(NETBEN_SMC, graph=F)]
df1_dt_1[, summ(NETBEN_SMP, graph=F)]
df1_dt_1[, summ(NETBEN_VASS, graph=F)] 
df1_dt_1[, summ(NETBEN_VAH, graph=F)] 
df1_dt_1[, summ(NETBEN_RVC, graph=F)] 
df1_dt_1[, summ(NETBEN_RCA, graph=F)] 
df1_dt_1[, summ(NETBEN_DC, graph=F)] 
df1_dt_1[, summ(NETBEN_AW, graph=F)] 
df1_dt_1[, summ(NETBEN_CCR, graph=F)] 
df1_dt_1[, summ(NETBEN_VR, graph=F)] 
```

Now it is time to produce a graph showing the net benefit of each of the 18 technologies, in a comparative way
```{r}


# FIRST WE CREATE THE WEIGHTED MEANS FOR EACH TECHNOLOGY

WeightedMeans <- lapply(df1_dt_1[ , c("NETBEN_MP", "NETBEN_BC", "NETBEN_PS", "NETBEN_WEL", "NETBEN_JOB", "NETBEN_CAN", "NETBEN_LOAN", "NETBEN_SMC", "NETBEN_SMP","NETBEN_VASS", "NETBEN_VAH", "NETBEN_RVC", "NETBEN_RCA", "NETBEN_DC", "NETBEN_AW", "NETBEN_CCR", "NETBEN_VR")], weighted.mean,  w = df1_dt_1$PV25_Weight, na.rm = T)

# Now we put this in the correct format

WeightedMeans_df <- data.frame(WeightedMeans)

WeightedMeans_dt <- as.data.table(WeightedMeans_df)

# transpose
t_WeightedMeans_dt <- transpose(WeightedMeans_dt)
t_WeightedMeans_dt[, VarNames := colnames(WeightedMeans_dt)]
colnames(t_WeightedMeans_dt)[1] <- "Net Benefit" 

t_WeightedMeans_df <- data.frame(t_WeightedMeans_dt)

#We re-code the names so they look nice in the graph

t_WeightedMeans_df_RC <- t_WeightedMeans_df %>%
  mutate("VarNames" = 
           dplyr::recode(VarNames, 
                         "NETBEN_MP" = "Face recognition: unblock",
                         "NETBEN_BC" = "Face recognition: border",
                         "NETBEN_PS" = "Face recognition: policing",
                         "NETBEN_WEL" = "Eligibility: welfare",
                         "NETBEN_JOB" = "Eligibility: jobs",
                         "NETBEN_CAN" = "Predict: risk cancer",
                         "NETBEN_LOAN" = "Predict: loan",
                         "NETBEN_SMC" = "Social media ads: product",
                         "NETBEN_SMP" = "Social media ads: political",
                         "NETBEN_VASS" = "Virtual assistant: speaker",
                         "NETBEN_VAH" = "Virtual assistsant: health",
                         "NETBEN_RVC" = "Robotics: vacuum",
                         "NETBEN_RCA" = "Robotics: care assistant",
                         "NETBEN_DC" = "Autonomous: cars",
                         "NETBEN_AW" = "Autonomous: weapons",
                         "NETBEN_CCR" = "Simulation: climate change",
                         "NETBEN_VR" = "Simulation: VR education"))

#And now we can plot


t_WeightedMeans_df_RC <- t_WeightedMeans_df_RC %>% mutate(VarNames = fct_reorder(VarNames, Net.Benefit))


p<-ggplot(data=t_WeightedMeans_df_RC, aes(x=Net.Benefit, y=VarNames)) +
  geom_bar(stat="identity") +
  ggtitle("Net benefit across AI applications") +
  xlab("Net benefit") + ylab("AI application") + scale_x_continuous(limits=c(-3, 3), n.breaks = 6) + 
  geom_col(fill = "#FF6A58") +
  theme(plot.title = element_text(hjust = 0.5))

p



```